We have 3 methods to do the cloud native initialization task：

1. Initialize with Helm pre-install hook and initialization containers

   If the hook depends on the service that will be installed, it will cause a deadlock.

2. Initialize with Helm post-install hook and initialization containers

   If the hook depends on the service that will be installed，and given wrong priority (hook weight), it will cause a deadlock.

   When we assign the correct priority to the hook, the hook executes serially, and the wait time will not be as long as necessary.

   If you let helm wait for the Pods in a ready state, deadlocks can occur. This is because helm is waiting before the post-hooks execution.

3. Initialize with Kubernetes job and initialization containers

   Without careful tuning of task priorities, it will complete the initialization work in the correct priority and shortest time.

   Since there is no helm to clean up, only the data required for initialization will remain in the cluster, and the Kubernetes environment will looks a bit messy, or need clean up manually.

The demo program requires the following prerequisites：

• Kubernetes environment
• The host address and port used to display the execution information
• Running netcat in the specified host address and port

This demo program contains 3 Microservice services, cm, nef and nidd, all of which depend on the database.

Let's assume that the startup time for these services and databases, database takes 10 seconds, cm takes 15 seconds, nidd takes 20 seconds. For the oneshot database initialization time, cm takes 10 seconds, nef takes 10 seconds, nidd takes 15 seconds.

Moreover, nef need oneshot initialization on cm, it takes 10 seconds。

1. Separate Helm charts works good.
2. Umbrella Helm charts can deadlock, not works.

1. Separate Helm charts works good.
2. Umbrella Helm charts works.
3. Umbrella Helm charts with improper hooks weight can deadlock, not works.
4. If you let helm wait for the Pods in a ready state, deadlocks can occur.

1. Separate Helm charts works good.
2. Umbrella Helm charts works good.
3. Kubernetes jobs can do better, it run initialization tasks concurrently, reducing initialization time.
4. Since no helm to clean up, it looks a bit messy, and add some confusion to the upgrade process.